关键词: 制造过程, 多重共线性, 局部线性嵌入, 特征选择

Abstract: In manufacturing processes, a large number of parameters were easily caused to have multicollinearity, which led to problems such as inaccurate prediction of quality indicators. To address these problems, a feature selection method for multicollinear parameters in the manufacturing processes was proposed based on LLE. Firstly, the multicollinearity of the manufacturing process parameters was diagnosed, and then the multicollinearity was eliminated by using the least absolute shrinkage and selection operator(LASSO) regression. Secondly, the LLE algorithm was used to perform feature selection on the parameters after LASSO regression to obtain independent feature spaces, and they were input into the whale optimization algorithm-support vector machine(WOA-SVM) model to verify the parameter feature selection effectiveness of the proposed algorithm. Finally, the effectiveness of the proposed method was verified through case analysis. The results show that compared with the original data, the proposed method may obtain more accurate prediction results under a lower-dimensional feature space, the correlation coefficient value is up to 0.9702, and the accuracy of feature selection increases by 24.989%.

Key words: manufacturing process, multicollinearity, local linear embedding(LLE), feature selection